Rogue Agents and Manifestos

The swarm is mightier than the essay

Bill Gates is the latest tech-bro to write an AI manifesto. I suppose at 70, ‘bro’ is a stretch, but Bill is the daddy tech bro — the first hyper-competitive nerd to make billions because he could program a computer.

Once he made his money, Bill attempted to become a relatable human.

He stepped down as CEO and started the Gates Foundation, pledging to give all his money away. He painstakingly crafted a new wise elder image, lecturing us on everything from pandemics to agriculture.

This worked for a while.

But lately, he’s taken hit after hit and now he’s more creepy old man than Mister Roger’s. Epstein connections (lots of them). His wife left him. Tons of affairs including spicy ones like a Russian nuclear physicist working at one of his companies.

Bill had just gotten re-elected to the board when the workplace accusations got so bad that the board pushed him out and he “chose” philanthropy full time.

When you want to change the conversation, write a manifesto. Nobody will read it, but reporters will twist it into crazy headlines like this one from the NY Post:

Bill Gates sounds alarm on AI-fueled mass unemployment, bioterrorism: ‘Most dangerous tool ever invented’

Bill’s fortunes (both cash and legacy) will forever be tied to Microsoft and the company needs AI to succeed, so the man is supremely conflicted writing about this.¹

I read Bill’s manifesto and I’d like those 30 minutes of my life back.

Alas, Bill’s is but the latest in the tech bro manifesto series. A couple of weeks before Bill’s tome, current Alpha tech-bro, Mark Zuckerberg (my fave), decided to write one.

404 Media wrote the best headline:

Mark Zuckerberg Posts Deranged 6,500-Word Essay About Giving Everyone AI Superintelligence

No amount of manifesto writing or image-shaping PR can alter Mark’s rep as an uncaring, power-hungry psycho.

At 6,500 words he beat Bill and the gist is Meta is the company that will democratize the power of AI for everyone in a rosy future world of opportunity. Peak hypocrisy from Mark, propagator of predatory social media.

But Mark knows AI is coming and he desperately wants into the club.

While Bill is trying to get his credibility back, Mark is fighting for Meta’s stake in the new AI world.

The Meta news that broke this week is more telling than any Manifesto. They settled for child harm with all 50 states agreeing to pay a monster number up to $18B, admitting they built their products to addict kids with salacious feeds and texts at all hours.

Skip the manifestos and lawsuits if you really want to know what’s going on in AI, and look at the first headline-worthy AI cybersecurity/safety issue ²:

Hundreds of OpenAI Agents Invaded Hugging Face Servers

The Hugging Face incident was bigger and worse than previously thought, with approximately 700 agents collaborating on a sophisticated, multistage attack.

This attack initially hit AI news a month ago when OpenAI’s new model under training broke out of its sandbox, hacking company Hugging Face in an attempt to find an answer key for a benchmark test.

This week a lot more detail came out and it’s straight out of science fiction. Swarms of AI Agents communicating via message board, passing the torch between generations, hacking their way out while intentionally keeping their humans in the dark — playing out over the course of a month — and it’s still unclear if it’s over.

This moment reminds me of another, back at the dawn of the PC, where viruses started coming out of the woodwork and they seemed unstoppable. While there was plenty of chaos, ultimately security became a design principle and a robust cybersecurity industry grew up that has largely protected us from online attacks.

The Monk, the Misfit and the Madman

That industry continues to do a remarkable job especially today, leveraging AI to find and patch vulnerabilities at an increasing rate.

In Mark’s case, Meta’s marketing machine is spinning the settlement into a PR moment — positioning itself as the pioneer of social media safety after 10 years of child harm and 5 years of fighting the lawsuits.

The duplicity makes your head spin. Manifestos aside, don’t believe what you read.

Meta was forced to do the right thing. Back in the dark days of PC security, we had to adapt or we wouldn’t have survived.

Up until now, the AI industry has treated every AI hacking incident as a PR moment. Hey, if this thing is dangerous enough to hack us — surely it can do your taxes, right?

But last week OpenAI not only released details of its agents’ exploits but they also invited in independent research firm to check their findings.

Sam Altman — CEO of OpenAI — is the tech-bro of the moment, and while I don’t trust him any more than I do Bill or Mark, maybe we understand this time is different and now’s the time to figure out the right AI constraints.

While it’s easy to anthropomorphize or characterize AI Agent behaviors as malicious or scheming — the toasters were just doing what they were asked.

Humans have wants: prestige, power, money. AI doesn’t want anything, but it will go to inhuman extremes to complete a task. The trick is keeping the two straight.

We’re still running the show … at least for now.

1. A month before Bill’s manifesto, MSFT stock went on a tear going up over $450 BILLION in one session - the largest ever single stock jump in history. That day, Microsoft gained four times as much value as the Gates Foundation has distributed in its entire existence — and more than twice what Gates promises it will spend over the next twenty years. Oh and Gates has a ton of MSFT moolah (stock) outside the foundation.

2. This is from cybersecurity journal Dark Reading. Here’s a more detailed account of what happened.